---
title: Pulumi Neo
url: /docs/ai/neo/
---
Pulumi Neo is Pulumi's own infrastructure agent, powered by Anthropic's Claude
family of models accessed via Amazon Bedrock. It ships with the Pulumi Agent
Skills catalog built in, and adds organizational context, policy guardrails,
human-in-the-loop approvals, and scheduled autonomous work.

Neo works on infrastructure the way a teammate would: it reads your
organization's live state in Pulumi Cloud, and depending on what you asked for,
it might answer a question, hand back an investigation, review a change, or
open a pull request against your IaC code.

## What Neo does

- **Answers questions about your infrastructure.** Ask which resources are
  running an outdated provider, why a stack failed, or what a change will do,
  and Neo searches your real state to answer. [Tasks](/docs/ai/neo/tasks/)
  covers how it investigates.
- **Proposes changes to your code.** When a task calls for changing
  infrastructure, Neo changes the IaC code rather than clicking in a console.
  The usual way it hands that back is a
  [pull request](/docs/ai/neo/pull-requests/) describing the problem, the
  resources affected, and a preview summary, so your branch protection rules
  and reviewers still gate what ships.
- **Runs previews before anything lands.** Neo can
  [run `pulumi preview`](/docs/ai/neo/running-previews/) to validate a change,
  surface policy violations, and show the resource diff.
- **Reviews your pull requests.** Neo
  [comments on PRs](/docs/ai/neo/code-reviews/) using what Pulumi Cloud knows
  about your running infrastructure, flagging issues inline. It never blocks
  the merge.
- **Takes on recurring maintenance.** Turn any task into
  [an automation](/docs/ai/neo/automations/) that runs on a schedule: provider
  freshness checks, encryption and backup audits, activity digests. Some
  automations report what they found; others open a pull request when there's
  work to do.
- **Applies Pulumi best practices.** Neo ships with the
  [Pulumi Agent Skills](/docs/ai/skills/) catalog built in, so it already knows
  proven workflows for components, ESC, migrations, and provider upgrades.

## Where to use Neo

Neo is one agent reachable from several places. Your permissions, approval
controls, and integrations carry over whichever surface you start from.

- **Pulumi Cloud console.** The Neo section in Pulumi Cloud is the main home for
  tasks, automations, and settings. [Get started](/docs/ai/neo/get-started/).
- **Your terminal.** Run [`pulumi neo`](/docs/ai/neo/pulumi-cli/) for an
  interactive session that inherits your local setup: the CLIs you've
  authenticated, your environment variables and kubeconfigs, and the project
  you're editing.
- **Slack.** Mention [`@Neo`](/docs/ai/neo/integrations/slack/) in any channel
  and it replies in the thread, so you can check stack state or investigate a
  failure without leaving the conversation.
- **Pull requests.** Neo reviews PRs automatically, and you can mention
  `@pulumi-neo` in a PR description, a review comment, or an issue to ask for
  something specific. See [code reviews](/docs/ai/neo/code-reviews/).
- **From another AI agent.** Claude Code, Codex, Cursor, and other agents can
  hand long-running infrastructure work to Neo through the
  [Pulumi MCP server](/docs/ai/mcp-server/) or the
  [Neo handoff skill](/docs/ai/skills/), then keep working while Neo runs.

Neo also reaches outward: [integrations](/docs/ai/neo/integrations/mcp/) connect
it to services like Datadog, PagerDuty, Linear, and Atlassian so it can pull in
the context a task needs.

## How much autonomy you give it

[Plan Mode](/docs/ai/neo/tasks/#plan-mode) makes Neo agree on a plan with you
before it acts, [task modes](/docs/ai/neo/tasks/#task-modes) control which steps
need your approval, and
[read-only mode](/docs/ai/neo/get-started/#read-only-mode) removes its ability
to change anything at all. Neo never has more access than you do, only less.